introduction - tu chemnitz€¦ · introduction the concept of 2-microlocal analysis or...
TRANSCRIPT
ATOMIC, MOLECULAR AND WAVELET DECOMPOSITION OFGENERALIZED 2-MICROLOCAL BESOV SPACES
HENNING KEMPKA
Abstract. We introduce generalized 2-microlocal Besov spaces and give char-acterizations in decomposition spaces by atoms, molecules and wavelets. Weapply the wavelet decomposition to prove that the 2-microlocal spaces areinvariant under the action of pseudodifferential operators of order 0.
1. Introduction
The concept of 2-microlocal analysis or 2-microlocal function spaces is due toJ.M. Bony (see [2]). It is an appropriate instrument to describe the local regularityand the oscillatory behavior of functions near singularities.The approach is Fourier-analytical using Littlewood-Paley-analysis of distributions.The theory has been elaborated and widely used in fractal analysis and signalprocessing by several authors. We refer to S. Jaffard ([8], [9]), Y. Meyer ([14]) andJ. Levy Vehel, S. Seuret ([12]). These works have been generalized in differentdirections by P. Andersson ([1]), H. Xu([15], [23]), and S. Moritoh, T. Yamada([16]).The main achievements are closely related to the use of wavelet methods and, as aconsequence, to wavelet characterizations of 2-microlocal spaces. Here, we intend togive a unified Fourier-analytical approach to generalized 2-microlocal Besov spacesand we are interested in characterizations by various decompositions.The paper is organized as follows. After recalling some preliminaries and notationin Section 2 we define the concept of 2-microlocal Besov spaces and repeat someresults from [10] in Section 3.In the main Sections 4 and 5 we present the characterizations of the 2-microlocalBesov spaces by decomposition in atoms, molecules and in wavelets. Only theproofs of the main theorems can be found in these sections, whereas the proofs ofthe more technical lemmas are shifted to the Appendix.Finally, in Section 6 we show that the 2-microlocal Besov spaces are invariant underthe action of certain Calderon-Zygmund operators introduced in [13] and the actionof pseudodifferential operators of order 0.
2. Preliminaries
As usual Rn denotes the n−dimensional Euclidean space, N is the collection ofall natural numbers and N0 = N ∪ {0}. Z and C stand for the sets of integers andcomplex numbers, respectively.The points of the Euclidian space x ∈ Rn are denoted by x = (x1, . . . , xn). If
Date: Version: October 23, 2008.2000 Mathematics Subject Classification. 42B35,42B20,47G30.Key words and phrases. 2-microlocal spaces,wavelet decomposition, Besov spaces.This work was supported by the DFG grant SCHM 969/7-1.
1
2 HENNING KEMPKA
β = (β1, . . . , βn) ∈ Nn0 is a multi-index, then its length is denoted by |β| = ∑n
j=1 βj .The derivatives Dβ = ∂|β|/∂β1
1 · · · ∂βnn have to be understood in the distributional
sense. We put xβ = xβ11 · · ·xβn
n .The Schwartz space S(Rn) is the space of all complex valued rapidly decreasinginfinitely differentiable functions on Rn. Its topology is generated by the norms
‖ϕ‖k,l = supx∈Rn
(1 + |x|)k∑
|β|≤l
|Dβϕ(x)| , k, l ∈ N0 .
A linear mapping f : S(Rn) → C is called a tempered distribution, if there exists aconstant c > 0 and k, l ∈ N0 such that
|f(ϕ)| ≤ c‖ϕ‖k,l
holds for all ϕ ∈ S(Rn). The collection of all such mappings is denoted by S ′(Rn).The Fourier transform is defined on both spaces S(Rn) and S ′(Rn) and is given by
(Ff)(ϕ) := f(Fϕ) , ϕ ∈ S(Rn) , f ∈ S ′(Rn)
where
Fϕ(ξ) :=1
(2π)n/2
∫
Rn
e−ix·ξ ϕ(x)dx .
Here x · ξ = x1ξ1 + · · · + xnξn stands for the inner product. The inverse Fouriertransform is denoted by F−1ϕ or ϕ∨ and we often write ϕ instead of Fϕ.We say a vector space E is a quasi-Banach space, if it is complete and quasi-normedby ‖ ·|E‖. That means ‖ ·|E‖ fulfills the norm conditions but the triangle inequalitychanges to
‖x + y|E‖ ≤ c ‖x|E‖+ c ‖y|E‖ , (1)
for c ≥ 1. If c = 1 in (1) then ‖ ·|E‖ is a norm and E is a Banach space. Asusual Lp(Rn) for 0 < p ≤ ∞ stands for the Lebesgue spaces on Rn normed by(quasi-normed if p < 1)
‖f |Lp(Rn)‖ =
∫
Rn
|f(x)|pdx
1/p
for 0 < p < ∞ and
‖f |L∞(Rn)‖ = ess-supx∈Rn
|f(x)| .
Let w be a positive measurable function. We define the weighted Lebesgue spaceLp(Rn, w) by ‖f |Lp(Rn, w)‖ = ‖wf |Lp(Rn)‖.For a sequence f = {fj}∞j=0 of vectors fj ∈ E the sequence spaces `q(E) for0 < q ≤ ∞ are normed by (quasi-normed if q < 1 or E is a quasi-Banach space)
‖f | `q(E)‖ =
∞∑
j=0
‖fj |E‖q
1/q
for 0 < q < ∞ and
‖f | `∞(E)‖ = supj∈N0
‖fj |E‖ .
If E = C then we shortly denote `q(C) by `q.The constant c stands for all unimportant constants. So the value of the constantc may change from one occurrence to another. By ak ∼ bk we mean that there aretwo constants c1, c2 > 0 such that c1ak ≤ bk ≤ c2ak for all admissible k.
DECOMPOSITIONS OF GENERALIZED 2-MICROLOCAL BESOV SPACES 3
3. Definitions and basic properties
In this section we present the Fourier analytical definition of generalized 2-microlocal Besov spaces Bs,mloc
pq (Rn, w) and we prove the basic properties in analogyto the classical Besov spaces. To this end we need smooth resolutions of unity andwe introduce admissible weight sequences w = {wj}j∈N0 .Definition 1 (Admissible weight sequence): Let α, α1, α2 ≥ 0. A sequence of non-negative measurable functions w = {wj}∞j=0 belongs to the class Wα
α1,α2if and only
if(i) There exists a constant C > 0 such that
0 < wj(x) ≤ Cwj(y)(1 + 2j |x− y|)α
for all j ∈ N0 and all x, y ∈ Rn.
(ii) For all j ∈ N0 we have
2−α1wj(x) ≤ wj+1(x) ≤ 2α2wj(x) for all x ∈ Rn.
Such a system {wj}∞j=0 ∈ Wαα1,α2
is called admissible weight sequence.A non-negative measurable function % is called an admissible weight function if
there exist constants α% ≥ 0 and C% > 0, such that
0 < %(x) ≤ C%%(y)(1 + |x− y|)α% holds for every x, y ∈ Rn. (2)
If w = {wj}∞j=0 is an admissible weight sequence, each wj is an admissible weightfunction, but in general the constant Cwj depends on j ∈ N0. If we use w ∈ Wα
α1,α2
without any restrictions, then α, α1, α2 ≥ 0 are arbitrary but fixed numbers.A fundamental example of an admissible weight sequence is given by the 2-microlocalweights. For a fixed nonempty set U ⊂ Rn and s′ ∈ R they are given by
wj(x) :=(1 + 2j dist(x, U)
)s′, (3)
where dist(x,U) = infz∈U |x− z| is the distance of x ∈ Rn from U .A special case is U = {x0} for x0 ∈ Rn. Then dist(U, x) = |x− x0| and we get thewell known 2-microlocal weights [9] treated by many authors
wj(x) = (1 + 2j |x− x0|)s′ for j ∈ N0. (4)
If U is an open subset of Rn, we get the weight sequence Moritoh and Yamada usedin [16].Another example of an admissible weight sequence is given for fixed s′ ∈ R by
wj(x) =(1 + 2jw(x)
)s′/β, (5)
where w : Rn → [0,∞) is a measurable function with the following properties:There are constants C1, C2 ≥ 1 and β ≥ 1 such that for all x, y ∈ Rn
0 ≤ w(x) ≤ C1w(y) + C2|x− y|β . (6)
As a special case we choose w : Rn → [0,∞) subadditiv, that is
0 ≤ w(x + y) ≤ c1 (w(x) + w(y)) and in addition we need
w(x) ≤ c2|x|β for all x, y ∈ Rn and fixed c1, c2, β ≥ 1.
Thus we have (6) with C1 = c1 and C2 = c1c2 and we can define the admissibleweight sequence as in (5). For all examples of admissible weight sequences, consid-ered above, it is easy to show, that w ∈ Wα
α1,α2if |s′| ≤ α, max(0,−s′) ≤ α1 and
max(0, s′) ≤ α2.Next we define the resolution of unity.
4 HENNING KEMPKA
Definition 2 (Resolution of unity): A system ϕ = {ϕj}∞j=0 ⊂ S(Rn) belongs tothe class Φ(Rn) if and only if
(i)
supp ϕ0 ⊆ {x ∈ Rn : |x| ≤ 2} and for all j ∈ Nsuppϕj ⊆ {x ∈ Rn : 2j−1 ≤ |x| ≤ 2j+1} .
(ii) For each β ∈ Nn0 there exist constants cβ > 0 such that
2j|β| supx∈Rn
|Dβϕj(x)| ≤ cβ holds for all j ∈ N0.
(iii) For all x ∈ Rn we have
∞∑
j=0
ϕj(x) = 1 .
Remark 1: Such a resolution of unity can easily be constructed. Consider thefollowing example. Let ϕ0 ∈ S(Rn) with ϕ0(x) = 1 for |x| ≤ 1 and supp ϕ0 ⊆ {x ∈Rn : |x| ≤ 2}. For j ≥ 1 we define
ϕ(x) = ϕ0(x)− ϕ0(2x) and ϕj(x) = ϕ(2−jx) .
Now it is obvious that ϕ = {ϕj}j∈N0 ∈ Φ(Rn).
Definition 3: Let {ϕj}j∈N0∈ Φ(Rn) be a resolution of unity and let w = {wj}j∈N0 ∈
Wαα1,α2
. Further, let 0 < p ≤ ∞, 0 < q ≤ ∞ and s ∈ R. Then we define
Bs,mlocpq (Rn, w) =
{f ∈ S ′(Rn) :
∥∥f |Bs,mlocpq (Rn, w)
∥∥ϕ
< ∞}
, where
∥∥f |Bs,mlocpq (Rn, w)
∥∥ϕ
=
∞∑
j=0
2jsq∥∥∥wj(ϕj f)∨
∣∣∣ Lp(Rn)∥∥∥
q
1/q
,
with the usual modifications if q is equal to infinity.One recognizes immediately that for wj(x) = 1 for all x ∈ Rn and all j ∈ N0 we
get the usual Besov spaces, see [18]. If one defines the admissible weight sequenceas wj(x) = %(x) for each j ∈ N0 and % being an admissible weight satisfying (2),we obtain weighted Besov spaces, see [4, Chapter 4].Applying a Fourier multiplier theorem for weighted Lebesgue spaces of entire an-alytic functions as in Subsection 1.7.5 in [17] we can prove that Definition 3 isindependent of the chosen resolution of unity {ϕj}j∈N0
∈ Φ(Rn), in the sense ofequivalent quasi-norms. To this end, we can suppress the index ϕ in the notationof the norm.Moreover, as in [18, Subsection 2.3.3] we can prove that Bs,mloc
pq (Rn, w) is a quasi-Banach space for all s ∈ R and 0 < p, q ≤ ∞ and even a Banach space in the casep, q ≥ 1. Also, we have for all admissible weight sequences and all s, p and q
S(Rn) ⊂ Bs,mlocpq (Rn,w) ⊂ S ′(Rn) ,
where S(Rn) is dense in Bs,mlocpq (Rn, w) for 0 < p, q < ∞. See [10] for details. Now,
let us recall some results from [10] we need later on.
DECOMPOSITIONS OF GENERALIZED 2-MICROLOCAL BESOV SPACES 5
Proposition 1: Let s ∈ R, w = {wj}j∈N0 ∈ Wαα1,α2
, 0 < p, q ≤ ∞ and let m ∈ N0.Then ∑
|β|≤m
∥∥Dβf∣∣ Bs−m,mloc
pq (Rn, w)∥∥ and
∥∥f |Bs,mlocpq (Rn, w)
∥∥
are equivalent quasi-norms on Bs,mlocpq (Rn, w).
The spaces Bs,mlocpq (Rn,w) have the lift property. To be precise, let us define the
lift operator Iσ by
Iσ : f 7→(〈ξ〉σ f
)∨(7)
where 〈ξ〉 = (1 + |ξ|2)1/2.Proposition 2: Let s, σ ∈ R and w = {wj}j∈N0 ∈ Wα
α1,α2. Moreover, let 0 <
p ≤ ∞ and 0 < q ≤ ∞. Then Iσ maps Bs,mlocpq (Rn,w) isomorphically onto
Bs−σ,mlocpq (Rn, w). Further,
∥∥Iσf |Bs−σ,mlocpq (Rn, w)
∥∥ is an equivalent quasi-normon Bs,mloc
pq (Rn, w).
The next result is the characterization of Bs,mlocpq (Rn, w) by local means. There-
fore, we define for {ψk} ∈ S(Rn), f ∈ S ′(Rn) and a > 0 the Peetre maximaloperator by
(ψ∗kf)a(x) = supy∈Rn
|(ψk ∗ f)(y)|1 + |2k(y − x)|a (k ∈ N0, x ∈ Rn) . (8)
Proposition 3: Let w = {wk}k∈N0 ∈ Wαα1,α2
, 0 < p, q ≤ ∞ and let s, a ∈ R,R ∈ N0 with a > n
p + α and R > s + α2. Let ψ0, ψ1 ∈ S(Rn) such that
Dβψ∨1 (0) = 0 , for 0 ≤ |β| < R , (9)
and
|ψ∨0 (x)| > 0 on {x ∈ Rn : |x| < ε} (10)
|ψ∨1 (x)| > 0 on {x ∈ Rn : ε/2 < |x| < 2ε} (11)
for some ε > 0. Further, let ψk(x) = 2(k−1)nψ1(2k−1x), then∥∥f |Bs,mloc
pq (Rn,w)∥∥ ∼
∥∥2ks(ψk ∗ f)wk
∣∣ lq(Lp)∥∥ ∼
∥∥2ks(ψ∗kf)awk
∣∣ lq(Lp)∥∥
holds for all f ∈ S ′(Rn).If R = 0, then we do not need any moment conditions (9) on ψ1.
Now, we present an elementary embedding in the scale of Bs,mlocpq (Rn,w).
Proposition 4: Let s ∈ R and w,% ∈ Wαα1,α2
with wj(x)%j(x) ≤ c for all j ∈ N0 and
x ∈ Rn. If 0 < p ≤ ∞, 0 < q1 ≤ ∞, 0 < q2 ≤ ∞ and ε > 0, then
Bs,mlocpq1
(Rn,%) ↪→ Bs−ε,mlocpq2
(Rn, w) .
For further embeddings in the general scale of Bs,mlocpq (Rn, w) and also in the
special scale Bs,s′pq (Rn, U) with the 2-microlocal weights wj(x) = (1+2j dist(x,U))s′
we refer to [10].
4. Decomposition by Atoms and Molecules
In this chapter we present two decomposition theorems. We characterize thespaces Bs,mloc
pq (Rn,w) via decompositions by atoms and molecules. First we intro-duce the basic notation.
6 HENNING KEMPKA
4.1. Sequence Spaces. First of all, we define for ν ∈ N0 and m ∈ Zn the closedcube Qνm with center in 2−νm and with sides parallel to the axes and length 2−ν .By χνm we denote the characteristic function of the cube Qνm, defined by
χνm(x) = χQνm(x) =
{1 , if x ∈ Qνm
0 , if x /∈ Qνm .
Definition 4: Let w = {wk}k∈N0 ∈ Wαα1,α2
, s ∈ R and let 0 < p, q ≤ ∞. Then forall complex valued sequences
λ = {λνm ∈ C : ν ∈ N0,m ∈ Zn}we define
bs,mlocpq (w) =
λ :∥∥λ| bs,mloc
pq (w)∥∥ :=
∞∑ν=0
2ν(s−np )q
( ∑
m∈Zn
wpν(2−νm)|λνm|p
)q/p
1/q
with the usual modifications if p or q are equal to infinity.Remark 2: Observe, that
∥∥λ| bs,mlocpq (w)
∥∥ ∼( ∞∑
ν=0
2νsq
∥∥∥∥∥∑
m∈Zn
λνmχνm(·)wν(·)∣∣∣∣∣ Lp(Rn)
∥∥∥∥∥
q)1/q
. (12)
4.2. Atoms and Molecules. Atoms are the building blocks for atomic decompo-sitions.Definition 5: Let K, L ∈ N0 and let γ > 1. A K-times continuous differentiablefunction a ∈ CK(Rn) is called [K,L]-atom centered at Qνm, ν ∈ N0 and m ∈ Zn,if
supp a ⊆ γQνm , (13)
|Dβa(x)| ≤ 2|β|ν , for 0 ≤ |β| ≤ K (14)
and if∫
Rn
xβa(x)dx = 0 for 0 ≤ |β| < L and ν ≥ 1 . (15)
If an atom a is centered at Qνm, that means if it fulfills (13), then we denote itby aνm. We recall the definition xβ = xβ1
1 · · ·xβnn and point out that for ν = 0 or
L = 0 there are no moment conditions (15) required.Definition 6: Let K,L ∈ N0 and let M > 0. A K-times continuous differentiablefunction µ ∈ CK(Rn) is called [K, L, M ]-molecule concentrated in Qνm, if for someν ∈ N0 and m ∈ Zn
|Dβµ(x)| ≤ 2|β|ν(1 + 2ν |x− 2−νm|)−M , for 0 ≤ |β| ≤ K (16)
and∫
Rn
xβµ(x)dx = 0 if 0 ≤ |β| < L and ν ≥ 1 . (17)
Remark 3: (a) For L = 0 or ν = 0 there are no moment conditions (17)required. If a molecule is concentrated in Qνm, that means it satisfies (16),then it is denoted by µνm.
DECOMPOSITIONS OF GENERALIZED 2-MICROLOCAL BESOV SPACES 7
(b) If aνm is a [K, L]-atom then it is a [K, L,M ]-molecule for every M > 0.
First, we show the convergence of the molecular decomposition. To this end, weintroduce the number σp, defined by σp = max(0, n(1/p− 1)).Lemma 1: Let w = {wk}k∈N0 ∈ Wα
α1,α2and let 0 < p, q ≤ ∞, s ∈ R. Further-
more, let K,L ∈ N0 and M > 0 with
L > σp − s + α1 , K ∈ N0 and M large enough .
If λ ∈ bs,mlocpq (w) and {µνm}νN0,m∈Zn are [K,L, M ]-molecules concentrated in Qνm,
then the sum∞∑
ν=0
∑
m∈Zn
λνmµνm(x) (18)
converges in S ′(Rn).Remark 4: A sufficient condition on M is M > L + 2α + 2n.
We also need a representation formula of Calderon type. The proof can be foundin [5, Theorem 2.6].Lemma 2: Let {ϕj}j∈N0 ∈ Φ(Rn) be a resolution of unity and let M ∈ N. Thenthere exist functions θ0, θ ∈ S(Rn) with:
supp θ0, supp θ ⊆ {x ∈ Rn : |x| ≤ 1} , (19)
|θ0(ξ)| ≥ c0 > 0 for |ξ| ≤ 2 , (20)
|θ(ξ)| ≥ c > 0 for12≤ |ξ| ≤ 2 , (21)
∫
Rn
xγθ(x)dx = 0 for 0 ≤ |γ| ≤ M (22)
such that
θ0(ξ)ψ0(ξ) +∞∑
j=1
θ(2−jξ)ψ(2−jξ) = 1 , for all ξ ∈ Rn , (23)
where the functions ψ0, ψ ∈ S(Rn) are defined by
ψ0(ξ) =ϕ0(ξ)
θ0(ξ)and ψ(ξ) =
ϕ1(2ξ)
θ(ξ). (24)
We have already seen that the sum in (18) converges in S ′(Rn) under the con-ditions of Lemma 1. Now we come to the atomic decomposition theorem.Theorem 1: Let w = {wk}k∈N0 ∈ Wα
α1,α2, s ∈ R and let 0 < p, q ≤ ∞. Further-
more, let K,L ∈ N0 with
K > s + α2 and L > σp − s + α1 .
For each f ∈ Bs,mlocpq (Rn, w) there exist λ ∈ bs,mloc
pq (w) and [K, L]-atoms {aνm}centered at Qνm such that the representation
f =∞∑
ν=0
∑
m∈Zn
λνmaνm convergence being in S ′(Rn), (25)
8 HENNING KEMPKA
holds. Moreover,∥∥λ| bs,mloc
pq (w)∥∥ ≤ c
∥∥f |Bs,mlocpq (Rn, w)
∥∥ ,
where the constant c is universal for all f ∈ Bs,mlocpq (Rn,w).
Proof. Since [K, L]-atoms are [K, L, M ]-molecules for every M > 0 the convergencein S ′(Rn) has already been stated in Lemma 1.The proof relies on the method used in the proof of [7, Theorem 5.11]. We useLemma 2 with M = L − 1, the functions θ0, θ ∈ S(Rn) with the properties (19)-(23) and the functions ψ0, ψ ∈ S(Rn) with (24). Let f ∈ Bs,mloc
pq (Rn,w), then weget from the Lemma 2
f = f ∗ θ0 ∗ ψ0 +∞∑
ν=1
2νnθ(2ν ·) ∗ ψν ∗ f ,
where ψν(·) = 2νnψ(2ν ·). Now, splitting the integration with respect to the cubesQνm we derive
f(x) =∑
m∈Zn
∫
Q0m
θ0(x− y)(ψ0 ∗ f)(y)dy +∞∑
ν=1
∑
m∈Zn
2νn
∫
Qνm
θ(2ν(x− y))(ψν ∗ f)(y)dy .
We define for each ν ∈ N and all m ∈ Zn
λνm = Cθ supy∈Qνm
|(ψν ∗ f)(y)| , (26)
where Cθ = max{sup|x|≤1 |Dβθ(x)| : |β| ≤ K}. If λνm 6= 0, then we define
aνm(x) =1
λνm2νn
∫
Qνm
θ(2ν(x− y))(ψν ∗ f)(y)dy , (27)
otherwise we set aνm(x) = 0. The a0m atoms and λ0m are defined similarly usingθ0 and ψ0. Clearly, (25) and the properties of θ0, ψ0, θ and ψν ensure that aνm
are [K, L]-atoms. It remains to prove, that there exists a constant c such that∥∥λ| bs,mlocpq (w)
∥∥ ≤ c∥∥f |Bs,mloc
pq (Rn, w)∥∥. We have for fixed ν ∈ N0 and a > n
p + α∑
m∈Zn
wν(x)λνmχνm(x) ≤ c∑
m∈Zn
wν(x) supy∈Qνm
|(ψν ∗ f)(y)|χνm(x)
≤ c′wν(x) sup|z|≤c2−ν
|(ψν ∗ f)(x− z)|1 + |2νz|a (1 + |2νz|a)
≤ c′′wν(x)(ψ∗νf)a(x) ,
since |x− y| ≤ c2−ν for x, y ∈ Qνm and∑
m∈Zn χνm(x) = 1. Here, (ψ∗νf)a denotesthe Peetre maximal operator, defined in (8). Therefore, we have using (12)
∥∥λ| bs,mlocpq (w)
∥∥ ≤ c
( ∞∑ν=0
2νsq ‖wν(x)(ψ∗νf)a(x)|Lp(Rn)‖q
)1/q
. (28)
Since ψ0 ∈ S(Rn) and ψ ∈ S(Rn) are two kernels which fulfill the moment con-ditions (9) and the Tauberian conditions (10) and (11), we can use Proposition 3with a > n
p + α and derive from (28)∥∥λ| bs,mloc
pq (w)∥∥ ≤ c
∥∥f |Bs,mlocpq (Rn, w)
∥∥ .
¤
DECOMPOSITIONS OF GENERALIZED 2-MICROLOCAL BESOV SPACES 9
To prove the converse direction of the atomic decomposition, we take the moregeneral molecules in the above sense. To this end, we need two technical lemmasas used in [5]. We want to estimate the size of ϕ∨j ∗ µνm, where {ϕj} ∈ Φ(Rn) andµνm is a [K, L, M ]-molecule concentrated in Qνm. We use the resolution of unityintroduced in Remark 1. We recall Lemma 3.3 in [5].Lemma 3: Let {ϕj}j∈N0 ∈ Φ(Rn) be a resolution of unity and {µνm}ν∈N0,m∈Zn
are [K, L, M ]-molecules. Then we have for all x ∈ Rn
|(ϕ∨j ∗ µνm)(x)| ≤ c2−(ν−j)(L+n)(1 + 2j |x− 2−νm|)L+n−M , for j ≤ ν (29)
and
|(ϕ∨j ∗ µνm)(x)| ≤ c2−(j−ν)K(1 + 2j |x− 2−νm|)−M , for j ≥ ν. (30)
The next lemma is analogous to Lemma 3.4 in [5] and the proof is in the Ap-pendix.Lemma 4: Let 1 ≤ p ≤ ∞, w ∈ Wα
α1,α2and let
F (x) =∑
m∈Zn
λνmfνm ,
where λνm ∈ C and fνm(x) = (1 + 2j |x− 2−νm|)−R for R > α + n.Then we have for j ≤ ν
‖F |Lp(Rn, wj)‖ ≤ c2−ν np 2(ν−j)(α1+n)
( ∑
m∈Zn
|λνm|pwpν(2−νm)
)1/p
and for j ≥ ν
‖F |Lp(Rn, wj)‖ ≤ c2−ν np 2(j−ν)α2
( ∑
m∈Zn
|λνm|pwpν(2−νm)
)1/p
.
Finally, we can state the converse direction of the decomposition theorem bymolecules.Theorem 2: Let w ∈ Wα
α1,α2, s ∈ R and let 0 < p, q ≤ ∞. Further, let K, L ∈ N0
with
K > s + α2 , L > σp − s + α1
and M > 0 large enough (M > L + n(max(1, 1/p) + 1) + 2α). If {µνm} are[K,L, M ]-molecules and λ = {λνm} ∈ bs,mloc
pq (w), then
f =∞∑
ν=0
∑
m∈Zn
λνmµνm , convergence being in S ′(Rn), (31)
is an element of Bs,mlocpq (Rn,w) and
∥∥f |Bs,mlocpq (Rn, w)
∥∥ ≤ c∥∥λ| bs,mloc
pq (w)∥∥ .
10 HENNING KEMPKA
Proof. We have the representation of f ∈ S ′(Rn) by (31) and we know by Lemma1 that this representation converges. Now, we estimate the norm of f
∥∥f |Bs,mlocpq (Rn, w)
∥∥ =
∞∑
j=0
2jsq∥∥∥ (ϕj f)∨wj
∣∣∣ Lp(Rn)∥∥∥
q
1/q
≤ c
∞∑
j=0
2jsq
∥∥∥∥∥∞∑
ν=0
∑
m∈Zn
λνm(ϕ∨j ∗ µνm)wj
∣∣∣∣∣ Lp(Rn)
∥∥∥∥∥
q
1/q
≤ c
∞∑
j=0
2jsq
∥∥∥∥∥j∑
ν=0
∑
m∈Zn
. . .
∣∣∣∣∣ Lp(Rn)
∥∥∥∥∥
q
1/q
+ c
∞∑
j=0
2jsq
∥∥∥∥∥∥
∞∑
ν=j+1
∑
m∈Zn
. . .
∣∣∣∣∣∣Lp(Rn)
∥∥∥∥∥∥
q
1/q
.
(32)
Let p ≥ 1. We estimate the first Lp Norm in (32). We use (30) and we derive fromLemma 4 with R = M > n + α
Ij :=
∥∥∥∥∥j∑
ν=0
∑
m∈Zn
λνm(ϕ∨j ∗ µνm)wj
∣∣∣∣∣ Lp(Rn)
∥∥∥∥∥
≤ c
j∑ν=0
2−ν np 2−(j−ν)(K−α2)
( ∑
m∈Zn
|λνm|pwpν(2−νm)
)1/p
.
For the second term in (32) we use (29) and Lemma 4 with R = M −L−n > n+αand get
IIj :=
∥∥∥∥∥∥
∞∑
ν=j+1
∑
m∈Zn
λνm(ϕ∨j ∗ µνm)wj
∣∣∣∣∣∣Lp(Rn)
∥∥∥∥∥∥
≤ c
∞∑
ν=j+1
2−ν np 2−(ν−j)(L−α1)
( ∑
m∈Zn
|λνm|pwpν(2−νm)
)1/p
.
Now, we consider the case 0 < p < 1. We use the embedding `p ↪→ `1 and obtainfor the first term in (32)
Ij ≤
∫
Rn
(j∑
ν=0
∑
m∈Zn
|λνm||(ϕ∨j ∗ µνm)(x)|wj(x)
)p
dx
1/p
≤ c
j∑ν=0
∑
m∈Zn
|λνm|p∫
Rn
|(ϕ∨j ∗ µνm)(x)|pwpj (x)dx
1/p
By applying (30) and using the properties of the weight sequence, we estimate∫
Rn
|(ϕ∨j ∗ µνm)(x)|pwpj (x)dx ≤ c2−(j−ν)(K−α2)p×
×∫
Rn
(1 + 2ν |x− 2−νm|)(−M+α)pwpν(2−νm)dx
≤ c′2−(j−ν)(K−α2)p2−νnwpν(2−νm) , for M >
n
p+ α.
DECOMPOSITIONS OF GENERALIZED 2-MICROLOCAL BESOV SPACES 11
So, we get for the first Lp norm
Ij ≤ c
(j∑
ν=0
2−νn2−(j−ν)(K−α2)p∑
m∈Zn
|λνm|pwpν(2−νm)
)1/p
.
By a similar calculation we obtain the second estimate (M > np + L + n + α)
IIj ≤ c
∞∑
ν=j+1
2−νn2−(ν−j)(L−σp−α1)p∑
m∈Zn
|λνm|pwpν(2−νm)
1/p
.
We denote p := min(1, p) and t := min(1, p, q). We can rewrite our results for thefirst term as
Ij ≤ c
j∑ν=0
2−ν np p2−(j−ν)(K−α2)p
( ∑
m∈Zn
|λνm|pwpν(2−νm)
)p/p
1/p
,
for all 0 < p, q ≤ ∞. Finally, we conclude with that notation and `t/p ↪→ `1
∥∥2jsIj
∣∣ `q
∥∥t ≤ c
∥∥∥∥∥∥
j∑ν=0
2ν(s−np )t2−(j−ν)(K−s−α2)t
( ∑
m∈Zn
|λνm|pwpν(2−νm)
)t/p∣∣∣∣∣∣`q/t
∥∥∥∥∥∥,
and Young’s inequality gives us
≤ c′
∞∑ν=0
2ν(s−np )q
( ∑
m∈Zn
|λνm|pwpν(2−νm)
)q/p
t/q
∞∑
j=0
2−jζt
= c′′∥∥λ| bs,mloc
pq (w)∥∥t
, where ζ := K − s− α2 > 0.
With the same notation a similar estimate can be achieved for∥∥2jsIIj
∣∣ `q
∥∥t. Hereone has to use ζ := L− σp + s− α1 > 0 and this finishes the proof. ¤
For every M > 0 every [K, L] atom is a [K, L, M ] molecule. So we get an easycorollary for the atomic decomposition.Corollary 1: Let w = {wk}k∈N0 ∈ Wα
α1,α2and let 0 < p, q ≤ ∞, s ∈ R. Further-
more, let K,L ∈ N0 with
K > s + α2 and L > σp − s + α1 .
(i) If λ ∈ bs,mlocpq (w) and {aνm}ν∈N0,m∈Zn are [K, L]-atoms centered at Qνm,
then
f =∞∑
ν=0
∑
m∈Zn
λνmaνm (33)
belongs to the space Bs,mlocpq (Rn, w) and there exists a constant c > 0 with
∥∥f |Bs,mlocpq (Rn, w)
∥∥ ≤ c∥∥λ| bs,mloc
pq (w)∥∥ .
The constant c is universal for all λ and aνm.
12 HENNING KEMPKA
(ii) For each f ∈ Bs,mlocpq (Rn, w) there exist λ ∈ bs,mloc
pq (w) and [K,L]-atomscentered at Qνm such that f can be represented as in (33) (convergencebeing in S ′(Rn)), where
∥∥λ| bs,mlocpq (w)
∥∥ ≤ c∥∥f |Bs,mloc
pq (Rn,w)∥∥
and the constant c is universal for all f ∈ Bs,mlocpq (Rn, w).
5. Wavelet decomposition
In this section we describe the characterization of Bs,mlocpq (Rn, w) by a decom-
position in wavelets. We follow closely the ideas expressed in [19], [20] and [11].
5.1. Preliminaries. First of all, we recall some results from wavelet theory.Proposition 5: (i) There are a real scaling function ψF ∈ S(R) and a real
associated wavelet ψM ∈ S(R) such that their Fourier transforms have com-pact supports, ψF (0) = (2π)−1/2 and
supp ψM ⊆[−8
3π,−2
3π
]∪
[23π,
83π
].
(ii) For any k ∈ N there are a real compactly supported scaling function ψF ∈Ck(R) and a real compactly supported associated wavelet ψM ∈ Ck(R) suchthat ψF (0) = (2π)−1/2 and
∫
R
xlψM (x)dx = 0 for all l ∈ {0, . . . , k − 1} .
In both cases we have, that {ψνm : ν ∈ N0,m ∈ Z} is an orthonormal basis inL2(R), where
ψνm(t) :=
{ψF (t−m), if ν = 0,m ∈ Z2
ν−12 ψM (2ν−1t−m), if ν ∈ N, m ∈ Z
and the functions ψM , ψF are according to (i) or (ii).This Proposition is taken over from [21, Theorem 1.61]. The wavelets in the first
part are called Meyer wavelets. They do not have a compact support but they arerapidly decaying functions (ψF , ψM ∈ S(R)) and ψM has infinitely many momentconditions. The wavelets from the second part are called Daubechies wavelets.Here the functions ψM , ψF do have compact support, but they have only limitedsmoothness. Both types of wavelets are well described in [22], chapters 3 and 4.This orthonormal basis can be generalized to the n-dimensional case by a tensorproduct procedure. We take over the notation from [21, Subsection 4.2.1] withl = 0. Let ψM , ψF be the Meyer or Daubechies wavelets described above. Now, wedefine
G0 = {F, M}n and Gν = {F, M}n∗ if ν ≥ 1 ,
where the * indicates, that at least one Gi of G = (G1, . . . , Gn) ∈ {F,M}n∗ mustbe an M . It is clear from the definition, that the cardinal number of {F, M}n∗ is2n − 1. Let for x ∈ R
ΨνGm(x) = 2v n
2
n∏r=1
ψGr (2νxr −mr) , (34)
DECOMPOSITIONS OF GENERALIZED 2-MICROLOCAL BESOV SPACES 13
where G ∈ Gν , m ∈ Zn and ν ∈ N0. Then {ΨνGm : ν ∈ N0, G ∈ Gν ,m ∈ Zn} is
an orthonormal basis in L2(Rn). Finally, we have to adjust the sequence spacesbs,mlocpq (w) to our situation.
Definition 7: Let w ∈ Wαα1,α2
, s ∈ R and 0 < p, q ≤ ∞. Then
bs,mlocpq (w) :=
{λ = {λν
Gm} ⊂ C :∥∥∥λ| bs,mloc
pq (w)∥∥∥ < ∞
}where
∥∥∥λ| bs,mlocpq (w)
∥∥∥ =
∞∑ν=0
2ν(s−np )q
∑
G∈Gν
( ∑
m∈Zn
|λνGm|pwp
ν(2−νm)
)q/p
1/q
.
To get the wavelet characterization we use local means with kernels which onlyhave limited smoothness and we use the molecular decomposition described in theprevious section. This idea goes back to [20], [11] and [7]. First, we recall the localmeans with kernel k
k(t, f)(x) = t−n
∫
Rn
k
(y − x
t
)f(y)dy .
With t = 2−j , x = 2−j l where j ∈ N0 and l ∈ Zn, one gets
k(2−j , f)(2−j l) = 2jn
∫
Rn
k(2jy − l)f(y)dy
=∫
Rn
kjl(y)f(y)dy (35)
= kjl(f) .
First, assume that the expression (35) makes sense, at least formally. Later on weshow that (35) is a dual pairing. Now, the usual properties for k are transferred tothe kernels kjl.
Definition 8: Let kjl(x) ∈ CA(Rn) with j ∈ N0 and l ∈ Zn be functions in Rn
with
|Dβkjl(x)| ≤ c2j|β|+jn(1 + 2j |x− 2−j l|)−C , |β| ≤ A ∈ N0, C > 0 ,
for all x ∈ Rn, j ∈ N0, l ∈ Zn, and∫
Rn
xβkjl(x)dx = 0 , |β| < B ∈ N0 ,
for j ≥ 1 and l ∈ Zn.It is clear from the definition, that {2−jnkjl} are [A, B,C] molecules.
5.2. Duality. In this subsection we show that the expression
k(f) =∫
Rn
k(y)f(y)dy (36)
makes sense as a dual pairing. Here, w ∈ Wαα1,α2
, f ∈ Bs,mlocpq (Rn, w) and the
function k : Rn 7→ C belongs to some weighted space of continuously differentiablefunctions Cu(Rn, κ), κ ≥ 0,
Cu(Rn, κ) := {f ∈ Cu(Rn) : (1 + |x|)κDβf(x) ∈ C(Rn) for all |β| ≤ u} ,
14 HENNING KEMPKA
normed by
‖f |Cu(Rn, κ)‖ = max|β|≤u
supx∈Rn
(1 + |x|)κ|Dβf(x)| .
We need to introduce the dual space of Bs,mlocpq (Rn, w). For s ∈ R and max(p, q) <
∞, we have that S(Rn) is dense in Bs,mlocpq (Rn, w). Therefore, a linear functional on
Bs,mlocpq (Rn, w) can be interpreted as an element of S ′(Rn). That means, g ∈ S ′(Rn)
belongs to the dual space (Bs,mlocpq (Rn,w))′ of the space Bs,mloc
pq (Rn, w), if and onlyif there exists a positive number c such that
|g(ϕ)| ≤ c∥∥ϕ|Bs,mloc
pq (Rn,w)∥∥ for all ϕ ∈ S(Rn).
We do not want to give a full treatment of duality theory in Besov spaces, but onecan modify the statements and proofs in Section 2.11 in [18] and one derives atleast
Bσp−s,mlocp′q′ (Rn,w−1) ↪→ (
Bs,mlocpq (Rn,w)
)′, (37)
where σp = n(1/p− 1)+ and
1p
+1p′
= 1 if 1 ≤ p < ∞ and p′ = ∞ if 0 < p < 1. (38)
Similarly for q′. It is easy to see, that w−1 = { 1wj}j∈N0 ∈ Wα
α2,α1for w ∈ Wα
α1,α2.
For max(p, q) = ∞ we get a similar result, where the right hand side of (37) isreplaced by (Bs,mloc
pq (Rn, w))′, where Bs,mlocpq (Rn, w) is the completion of S(Rn)
in Bs,mlocpq (Rn, w). In this case we only have the dual pairing (35) for the Meyer
wavelets, because they are elements of S(Rn) and for the Daubechies wavelets,because they have a compact support.In the case max(p, q) < ∞ we can extend this to more general wavelet bases. Thewavelets belong to Cu(Rn, κ) and we show under which conditions on u ∈ N0 andκ ≥ 0 we have
Cu(Rn, κ) ↪→ Bσp−s,mlocp′q′ (Rn, w−1) .
First of all, we show C0(Rn, κ) ↪→ B−α1,mlocp′∞ (Rn, w−1). We use our fixed resolution
of unity from Remark 1, the weight properties and Young’s inequality to see that∥∥∥f |B−α1,mlocp′∞ (Rn,w−1)
∥∥∥ = supj∈N0
2−α1j∥∥∥w−1
j (ϕj f)∨∣∣∣ Lp′(Rn)
∥∥∥
≤ c supj∈N0
∥∥w−10 (ϕ∨j ∗ f)
∣∣ Lp′(Rn)∥∥
≤ c′ supj∈N0
∥∥ (w−10 ϕ∨j ) ∗ ((1 + | · |)αf)
∣∣ Lp′(Rn)∥∥
≤ c′′ ‖ (1 + | · |)αf |Lp′(Rn)‖ supj∈N0
∥∥ϕ∨j∣∣ L1(Rn)
∥∥
≤ c′′′∥∥f |C0(Rn, κ)
∥∥ ,
where the last inequality comes from ϕj = ϕ(2−j ·) ∈ S(Rn) and that κ > np′ + α.
Let us assume now, that f ∈ Cu(Rn, κ), then by applying Proposition 1 we conclude
Cu(Rn, κ) ↪→ Bu−α1,mlocp′∞ (Rn, w−1) .
Finally, Proposition 4 gives us for u > σp − s + α1
Cu(Rn, κ) ↪→ Bσp−s,mlocp′q′ (Rn, w−1) . (39)
DECOMPOSITIONS OF GENERALIZED 2-MICROLOCAL BESOV SPACES 15
Because of (37) and (39) the equation (36) makes sense at least for u > σp−s+α1.Further, we mention that all functions {kjl} from Definition 8 with A ≥ u, B ∈ N0
arbitrary and C ≥ κ belong to the space Cu(Rn, κ). So we see that (35) is welldefined for A > σp − s + α1 and C ≥ α + n, but these conditions will always befulfilled in the following theorems.
5.3. Wavelet isomorphism. We want to use the molecular decomposition ob-tained in the last section. We assume, that {µνm} are [K, L, M ] molecules andthat the {kjl} are the above given functions from Definition 8.Before stating the theorem we have to prove some fundamental lemmas. First,we have to give estimates of the quantity | 〈µνm, kjl〉 |. The proofs go back to [6,Appendix B].Lemma 5: (i) Let ν ≥ j, M > A + n and L ≥ A, then
| 〈µνm, kjl〉 | ≤ c2−(ν−j)(A+n)(1 + 2j |2−νm− 2−j l|)−min(M−A−n,C) . (40)
(ii) Let ν ≤ j, C > K + n and B ≥ K, then
| 〈µνm, kjl〉 | ≤ c2−(j−ν)K(1 + 2ν |2−νm− 2−j l|)−min(M,C−K−n) . (41)
The next lemma is presented in [11, Lemma 7.4]. Nevertheless, we present aproof in the Appendix because our notation is a bit different.Lemma 6: Let 1 ≤ p ≤ ∞ and bm ∈ C.
(i) Let ν ≥ j and R > n. Then(∑
l∈Zn
( ∑
m∈Zn
|bm|(1 + 2j |2−j l − 2−νm|)−R
)p)1/p
≤ c2(ν−j) np′
( ∑
m∈Zn
|bm|p)1/p
,
where 1p + 1
p′ = 1.(ii) Let ν ≤ j and R > n. Then
(∑
l∈Zn
( ∑
m∈Zn
|bm|(1 + 2ν |2−j l − 2−νm|)−R
)p)1/p
≤ c2(j−ν) np
( ∑
m∈Zn
|bm|p)1/p
.
Now, we are ready to state the first theorem, which gives us one direction of thewavelet decomposition. We define k(f) = {kjl(f) : j ∈ N0, l ∈ Zn}.Theorem 3: Let w ∈ Wα
α1,α2, s ∈ R and 0 < p, q ≤ ∞. Further, let {kjl} be as in
Definition 8 with C > 0 large enough and A,B ∈ N0 with
A > σp − s + α1 , B > s + α2.
Then for some c > 0 and all f ∈ Bs,mlocpq (Rn, w) ,
∥∥k(f)| bs,mlocpq (w)
∥∥ ≤ c∥∥f |Bs,mloc
pq (Rn, w)∥∥ .
Proof. We assume f ∈ Bs,mlocpq (Rn, w) and we use the atomic decomposition theo-
rem (Corollary 1) to get
f =∞∑
ν=0
∑
m∈Zn
λνmaνm , (42)
where {aνm} are [K,L] atoms with K = B > s + α2 and L = A > σp− s + α1. Wewant to show∥∥k(f)| bs,mloc
pq (w)∥∥ ≤ c
∥∥λ| bs,mlocpq (w)
∥∥ ≤ c′∥∥f |Bs,mloc
pq (Rn,w)∥∥ , (43)
16 HENNING KEMPKA
where the last inequality comes from the atomic decomposition theorem. We recall,that atoms have compact support and that they are [K, L, M ] molecules for everyM > 0. We split (42) dependent on j ∈ N0 into
f = fj + f j =j∑
ν=0
· · ·+∞∑
ν=j+1
· · · and get
kjl(f) =∫
Rn
kjl(y)f(y)dy =∫
Rn
kjl(y)fj(y)dy +∫
Rn
kjl(y)f j(y)dy .
To shorten notation we define A(m, l) := (1 + min(2ν , 2j)|2−νm− 2−j l|).Let ν ≤ j and 1 ≤ p ≤ ∞, then we get by Minkowski’s inequality and Lemma 5
∑
l∈Zn
∣∣∣∣∣∣
∫
Rn
kjl(y)fj(y)dy
∣∣∣∣∣∣
p
wpj (2−j l)
1/p
≤j∑
ν=0
(∑
l∈Zn
( ∑
m∈Zn
|λνm|wj(2−j l)| 〈 kjl, aνm〉 |)p)1/p
≤ c
j∑ν=0
2−(j−ν)(B−α2)
(∑
l∈Zn
( ∑
m∈Zn
|λνm|wν(2−νm)A(m, l)−C+α
)p)1/p
(44)
≤ c′j∑
ν=0
2−(j−ν)(K−α2−np )
( ∑
m∈Zn
|λνm|pwpν(2−νm)
)1/p
, (45)
where in the last step we used Lemma 6 with R = C − α > n.In the case 0 < p < 1 we have also the estimate (44) and get by `p ↪→ `1
∑
l∈Zn
∣∣∣∣∣∣
∫
Rn
kjl(y)fj(y)dy
∣∣∣∣∣∣
p
wpj (2−j l)
1/p
≤ c
j∑ν=0
2−(j−ν)(B−α2)
( ∑
m∈Zn
|λνm|pwpν(2−νm)
∑
l∈Zn
A(m, l)−p(C−α)
)1/p
. (46)
A direct calculation shows, that∑
l∈Zn A(m, l)−p(C−α) ≤ c2(j−ν)n for C > np + α.
Therefore, we get from (45) and (46)
2j(s−np )
∑
l∈Zn
∣∣∣∣∣∣
∫
Rn
kjl(y)fj(y)dy
∣∣∣∣∣∣
p
wpj (2−j l)
1/p
≤ c
j∑ν=0
2ν(s−np )2−(j−ν)ζ
( ∑
m∈Zn
|λνm|pwpν(2−νm)
)1/p
, (47)
where ζ = B − s− α2 > 0.Let us consider the case ν > j and 1 ≤ p ≤ ∞. Then we derive similar to the first
DECOMPOSITIONS OF GENERALIZED 2-MICROLOCAL BESOV SPACES 17
case with Minkowski’s inequality and Lemma 5∑
l∈Zn
∣∣∣∣∣∣
∫
Rn
kjl(y)f j(y)dy
∣∣∣∣∣∣
p
wpj (2−j l)
1/p
≤∞∑
ν=j+1
(∑
l∈Zn
( ∑
m∈Zn
|λνm|wj(2−j l)| 〈 kjl, aνm〉 |)p)1/p
≤ c
∞∑
ν=j+1
2−(ν−j)(A+n−α1)
(∑
l∈Zn
( ∑
m∈Zn
|λνm|wν(2−νm)A(m, l)−C+A+n+α
)p)1/p
(48)
≤ c′∞∑
ν=j+1
2−(ν−j)(A−α1+np )
( ∑
m∈Zn
|λνm|pwpν(2−νm)
)1/p
,
where we used Lemma 6 with R = C −A− n− α > n in the last step. We obtainnow with ρ = A + s− α1 > 0
2j(s−np )
∑
l∈Zn
∣∣∣∣∣∣
∫
Rn
kjl(y)f j(y)dy
∣∣∣∣∣∣
p
wpj (2−j l)
1/p
≤ c
∞∑
ν=j+1
2ν(s−np )2−(ν−j)ρ
( ∑
m∈Zn
|λνm|pwpν(2−νm)
)1/p
. (49)
For 0 < p < 1 we have again (48), use `p ↪→ `1 and obtain∑
l∈Zn
∣∣∣∣∣∣
∫
Rn
kjl(y)f j(y)dy
∣∣∣∣∣∣
p
wpj (2−j l)
1/p
≤ c
∞∑
ν=j+1
2−(ν−j)(A+n−α1)
( ∑
m∈Zn
|λνm|pwpν(2−νm)
∑
l∈Zn
A(m, l)−p(C−A−n−α)
)1/p
.
The sum over l ∈ Zn is bounded by a constant for C > np + n + A + α. Hence, we
get
2j(s−np )
∑
l∈Zn
∣∣∣∣∣∣
∫
Rn
kjl(y)f j(y)dy
∣∣∣∣∣∣
p
wpj (2−j l)
1/p
≤ c
∞∑
ν=j+1
2ν(s−np )2−(ν−j)ρ
( ∑
m∈Zn
|λνm|pwpν(2−νm)
)1/p
, (50)
where ρ = A + s− α1 − (n/p− n) > 0.Now the result (43) can be obtained from (47), (49) and (50) by standard arguments.
¤
Remark 5: As shown in the proof it is enough to assume
C > A + n
(1 +
1min(1, p)
)+ α .
18 HENNING KEMPKA
We come to the wavelet decomposition theorem. Let us assume, that
ψM ∈ Ck(Rn) and ψF ∈ Ck(Rn) (51)
are the real compactly supported Daubechies wavelets from Proposition 5, with∫
Rn
xβψM (x)dx = 0 for |β| < k. (52)
By the tensor product procedure (34), we have, that {ΨνG,m : ν ∈ N0, G ∈ Gν and m ∈
Zn} is an orthonormal basis in L2(Rn).Before coming to the theorem we clarify the convergence of
∞∑ν=0
∑
G∈Gν
∑
m∈Zn
λνGm2−ν n
2 ΨνGm with λ ∈ bs,mloc
pq (w) . (53)
We say that a series converges unconditionally, if any rearrangement of the seriesalso converges to the same outcome. We know that {2−ν n
2 ΨνGm} are [k, k, M ]
molecules for every M > 0 and therefore we have the unconditional convergence of(53) in S ′(Rn) from Lemma 1 with k > σp − s + α1.Moreover, the following proof shows the unconditional convergence of (53) inBs,mloc
pq (Rn, w) for 0 < p < ∞ and 0 < q < ∞. If 0 < p < ∞ and 0 < q ≤ ∞then we have unconditional convergence in Bσ,mloc
pq (Rn, w) with σ < s. For general0 < p, q ≤ ∞ we get the unconditional convergence in Bσ,mloc
pq (Rn,%) where σ < s
and % is an admissible weight sequence with %ν(x)wν(x) → 0 for |x| → ∞.
Theorem 4: Let w ∈ Wαα1,α2
, s ∈ R, 0 < p, q ≤ ∞ and
k > max(σp − s + α1, s + α2) (54)
in (51) and (52). Let f ∈ S ′(Rn). Then f ∈ Bs,mlocpq (Rn, w) if, and only if, it can
be represented as
f =∞∑
ν=0
∑
G∈Gν
∑
m∈Zn
λνGm2−ν n
2 ΨνGm with λ ∈ bs,mloc
pq (w) , (55)
with unconditional convergence in S ′(Rn) and in any space Bσ,mlocpq (Rn,%) with
σ < s and %ν(x)wν(x) → 0 for |x| → ∞. The representation (55) is unique, we have
λνGm = λν
Gm(f) = 2ν n2 〈 f, Ψν
Gm〉 (56)
and
I : f 7→ {2ν n
2 〈 f, ΨνGm〉
}(57)
is an isomorphic map from Bs,mlocpq (Rn, w) onto bs,mloc
pq (w). Moreover, if in additionmax(p, q) < ∞ then {Ψν
Gm} is in unconditional basis in Bs,mlocpq (Rn, w).
Proof. First Step: Let f ∈ S ′(Rn) be given by (55). Then by the support propertieswe have that {2−ν n
2 ΨνGm} are [k, k, M ] molecules for every M > 0. From Theorem
2 and (54) we obtain f ∈ Bs,mlocpq (Rn, w) and
∥∥f |Bs,mlocpq (Rn, w)
∥∥ ≤ c∥∥∥λ| bs,mloc
pq (w)∥∥∥ (58)
with c > 0 independent of λ ∈ bs,mlocpq (w).
Second Step: Let f ∈ Bs,mlocpq (Rn, w) then we can apply Theorem 3 with kνm =
DECOMPOSITIONS OF GENERALIZED 2-MICROLOCAL BESOV SPACES 19
2ν n2 Ψν
Gm. Since all conditions on kνm are fulfilled by (54) and the compact supportof the wavelets we get
∥∥∥λ(f)| bs,mlocpq (w)
∥∥∥ ≤ c∥∥f |Bs,mloc
pq (Rn, w)∥∥ . (59)
Third Step: For max(p, q) < ∞ we get the unconditional convergence of (55) inBs,mloc
pq (Rn, w) by (58) and the properties of the sequence spaces bs,mlocpq (w).
Let p < ∞ and q = ∞, then we get the convergence in Bσ,mlocp∞ (Rn,w) for all σ < s
by using (58) again and Holder’s inequality.To obtain the convergence for p = ∞ we have to compensate the behavior at infinityby introducing a weaker weight sequence % with %ν(x)
wν(x) → 0 for |x| → ∞. Then weget the unconditional convergence in Bσ,mloc
∞q (Rn,%) with σ < s as in the previouscase.A simple example of such a weaker weight sequence is given for every ε > 0 by
%ν(x) = (1 + 2ν |x|)−εwν(x) (60)
which belongs to Wα+εα1+ε,α2
.Fourth Step: We want to prove now the uniqueness of the coefficients. We define
g =∞∑
ν=0
∑
G∈Gν
∑
m∈Zn
λνGm2−ν n
2 ΨνGm (61)
where λνGm is given by (56). We want to show that g = f , or
〈 g, ϕ〉 = 〈 f, ϕ〉 for every ϕ ∈ S(Rn).
From the first step we have g ∈ Bs,mlocpq (Rn, w). The third step tells us that (61)
converges at least in Bσ,mlocpq (Rn, %) for all σ < s and % is given by (60) for some
ε > 0. Since k > σp − s + α1 we can find σ < s and ε > 0 such that Ψν′G′m′ still
belongs to the dual space (Bσ,mlocpq (Rn, %))′ (that means k > σp−σ+α1+ε). Because
of the convergence in Bσ,mlocpq (Rn, %) ,the dual pairing and the orthonormality of
{ΨνGm} in L2(Rn) we get
⟨g, Ψν′
G′m′
⟩=
⟨f, Ψν′
G′m′
⟩. (62)
This holds also for finite linear combinations of Ψν′G′m′ . For a function ϕ ∈ S(Rn)
we have the unique L2(Rn) representation
ϕ =∑
ν,G,m
2−j n2 〈ϕ,Ψν
Gm〉ΨνGm .
Since S(Rn) is a subspace in every Besov space considered this representation con-verges in (Bσ,mloc
p,q (Rn, w))′ and we get by (62)
〈 g, ϕ〉 = 〈 f, ϕ〉 .
Final Step: Hence, f ∈ S ′(Rn) belongs to Bs,mlocpq (Rn,w) if and only if, it can be
represented by (55). This representation is unique with coefficients (56). By (58),(61), with g = f , and (59) it follows
∥∥∥λ(f)| bs,mlocpq (w)
∥∥∥ ∼∥∥f |Bs,mloc
pq (Rn, w)∥∥ .
20 HENNING KEMPKA
Hence I in (57) is an isomorphic map from Bs,mlocpq (Rn, w) into bs,mloc
pq (w). Itremains to prove that this map is onto. Let λ ∈ bs,mloc
pq (w). Then it follows by theabove considerations that
f =∑
ν,G,m
λνGm2−ν n
2 ΨνGm ∈ Bs,mloc
pq (Rn, w) .
By the same reasoning as in the fourth step this representation is unique andλν
Gm = λνGm(f). This proves that I is a map onto. ¤
To prove the wavelet decomposition with Daubechies wavelets we used only theatomic decomposition theorem.Now, we present a wavelet decomposition theorem with the help of Meyer wavelets,described in Proposition 5. We have ψM , ψF ∈ S(Rn) and we have infinitely manymoment conditions on ψM . But we lose the compact support property for thewavelets. Here we need to use our molecular decomposition. The proof is the sameas in Theorem 4. We use again our wavelets once as molecules and once as kernelsfrom Definition 8 where the technicalities turn out to be easier because A,B, C areinfinite.Theorem 5: Let {Ψν
Gm} be the Meyer wavelets according to Proposition 5. Fur-ther, let w ∈ Wα
α1,α2, s ∈ R, 0 < p, q ≤ ∞ and let f ∈ S ′(Rn). Then f ∈
Bs,mlocpq (Rn, w) if, and only if, it can be represented as
f =∞∑
ν=0
∑
G∈Gν
∑
m∈Zn
λνGm2−ν n
2 ΨνGm with λ ∈ bs,mloc
pq (w) , (63)
with unconditional convergence in S ′(Rn) and in any space Bσ,mlocpq (Rn,%) with
σ < s and %ν(x)wν(x) → 0 for |x| → ∞. The representation (63) is unique,
λνGm = λν
Gm(f) = 2ν n2 〈 f, Ψν
Gm〉and
I : f 7→ {2ν n
2 〈 f, ΨνGm〉
}
is an isomorphic map from Bs,mlocpq (Rn, w) onto bs,mloc
pq (w). Moreover, if in additionmax(p, q) < ∞ then {Ψν
Gm} is in unconditional basis in Bs,mlocpq (Rn, w).
Remark 6: The wavelet characterization of Bs,mlocpq (Rn, w) is not restricted to
the two wavelet systems presented in Proposition 5. The proof of Theorem 4 alsoapplies, at least for max(p, q) < ∞ due to duality, to all wavelet systems {Ψν
Gm}which satisfy that 2−ν n
2 {ΨνGm} are [K, K,M ] molecules with
K > max(σp − s + α1, s + α2) and M > K + n max(2, 1 + 1/p) + 2α , (64)
where the condition on M is not sharp.The proofs can easily be extended to biorthogonal wavelet bases (see [11] for details).
5.4. Wavelet decomposition of Bs,s′pq (Rn,U). For a bounded subset U ⊂ Rn
and s′ ∈ R we have the 2-microlocal weights
wν(x) = (1 + 2ν dist(x,U))s′ . (65)
We know that w ∈ Wαα1,α2
for |s′| ≤ α and −min(0, s′) ≤ α1, max(0, s′) ≤ α2.The spaces Bs,s′
pq (Rn, U) are defined as the spaces Bs,mlocpq (Rn,w) with the special
DECOMPOSITIONS OF GENERALIZED 2-MICROLOCAL BESOV SPACES 21
weight sequence w defined in (65). The corresponding sequence spaces are definedby the norm
∥∥∥λ| bs,s′pq (U)
∥∥∥ =
∞∑ν=0
2ν(s−n/p)q∑
G∈Gν
( ∑
m∈Zn
|λνGm|p(1 + 2ν dist(2−νm,U))ps′
)q/p
1/q
.
Now, we can adapt everything from Theorem 4 and a wavelet decomposition forthe 2-microlocal Besov spaces with respect to the Daubechies wavelets follows.Corollary 2: Let U ⊂ Rn bounded, s′ ∈ R and w as in (65). Further, let s ∈ R,0 < p, q ≤ ∞ and
k > max(σp − s−min(0, s′), s + max(0, s′)) (66)
in (51) and (52). Let f ∈ S ′(Rn). Then f ∈ Bs,s′pq (Rn, U) if, and only if, it can be
represented as
f =∞∑
ν=0
∑
G∈Gν
∑
m∈Zn
λνGm2−ν n
2 ΨνGm with λ ∈ bs,s′
pq (U) , (67)
with unconditional convergence in S ′(Rn) and in any space Bt,t′pq (Rn, U) with t < s
and t′ < s′. The representation (67) is unique,
λνGm = λν
Gm(f) = 2ν n2 〈 f, Ψν
Gm〉
and
I : f 7→ {2ν n
2 〈 f, ΨνGm〉
}
is an isomorphic map from Bs,s′pq (Rn, U) onto bs,s′
pq (U). Moreover, if in additionmax(p, q) < ∞ then {Ψν
Gm} is in unconditional basis in Bs,s′pq (Rn, U).
Let us say a few words about the convergence of (67). As in Theorem 4 we haveunconditional convergence in Bs,s′
pq (Rn, U) for max(p, q) < ∞ and arbitrary U ∈ Rn,not necessarily bounded. For 0 < p < ∞ and 0 < q ≤ ∞ we have unconditionalconvergence in Bt,s′
pq (Rn, U) with s > t and also arbitrary U . Only in the case ofp = ∞ we need as an additional assumption that U ⊂ Rn has to be bounded to getthe unconditional convergence in Bt,t′
pq (Rn, U) for all s > t and s′ > t′.
Remark 7: For p ≥ 1 we have σp = 0 in condition (66). Now we can rewrite thiscondition as
k > max(|s|, |s + s′|) .
This is almost the same condition
k > max(max(s, s + s′), max(−(s + n),−(s + s′)))
used in [14, p.64, p.67] in connection with the orthonormal Daubechies wavelets.
According to Remark 6 we can also give a characterization of Bs,s′pq (Rn, U) by
other wavelet bases {ΨνGm} (for example the Meyer wavelets), where {Ψν
Gm} are[K,K, M ] molecules with the condition (64) on K and M .
22 HENNING KEMPKA
6. Pseudodifferential Operators
In this section we show that the 2-microlocal spaces Bs,mlocpq (Rn, w) are invariant
under the action of classical pseudodifferential operators of order 0. The correspond-ing results have already been shown in [8] for Cs,s′
x0(Rn) and in [1] for F s,s′
p,q (Rn, x0),1 ≤ p, q ≤ ∞.As in [8] we introduce operators T belonging to the class Op(Mγ) of Calderon-Zygmund operators, which were introduced in [13].A linear and continuous operator T : S(Rn) → S ′(Rn) possesses by the Schwartzkernel Theorem a distribution kernel K in S ′(Rn × Rn) such that
〈Tf, g〉 = 〈K, f ⊗ g〉 for all f, g ∈ S(Rn).
This kernel K belongs to the class Oγ for γ > 0, if its restriction to the set {(x, y) ∈Rn × Rn : x 6= y} is continuous and satisfies
|DβK(x, y)| ≤ c|x− y|−n−|β| for |β| < γ
and, with γ = m + r, where m ∈ N0 and r ∈ (0, 1]
|DβK(x, y)−DβK(x′, y)| ≤ c|x− x′|r|x− y|n+γ
for |x− x′| ≤ 12|x− y|
|DβK(x, y)−DβK(x, y′)| ≤ c|y − y′|r|x− y|n+γ
for |y − y′| ≤ 12|x− y|
for |β| = m. Moreover, T has to be bounded on L2(Rn) and T (Xβ) = T ∗(Xβ) = 0for |β| ≤ γ (see [13] for definition of T (Xβ)). Then Op(Mγ) =
⋃γ′>γ Oγ′ .
The class Op(Mγ) can be characterized by decay conditions on the terms of thematrix (〈TΨλ,Ψλ′〉) with respect to a wavelet basis {Ψλ} where Ψλ = Ψν
Gm (see[13, Chapter 8]).Proposition 6: An operator T belongs to Op(Mγ)if and only if its matrix coeffi-cients satisfy
|⟨
TΨνGm, Ψj
Gl
⟩| ≤ c
2−|j−ν|(γ′+n/2)
1 + (j − ν)2
(1 +
|2−νm− 2−j l|2−ν + 2−j
)−n−γ′
, (68)
for some γ′ > γ.The condition in (68) can be reformulated as
2(ν−j)n/2∣∣∣⟨
TΨνGm, Ψj
Gl
⟩∣∣∣ ≤ c
{2−(ν−j)γ′(1 + 2j |2−j l − 2−νm|)−n−γ′ , for ν ≥ j
2−(j−ν)(n+γ′)(1 + 2ν |2−j l − 2−νm|)−n−γ′ , for ν ≤ j.
(69)
That means that our matrix is almost diagonal (in the meaning of [6]) and we caneasily adopt the proof of Theorem 3 with (69) instead of the Lemma 5.Theorem 6: Let w ∈ Wα
α1,α2, 0 < p, q ≤ ∞ and s ∈ R. If f belongs to
Bs,mlocpq (Rn, w) and T belongs to Op(Mγ) with
γ > max(s + α2, σp − s + α1, σp + α)
then Tf belongs to Bs,mlocpq (Rn, w).
If we keep the last Theorem in mind, we get the following interpretation. Theposition of points of regularity of a function is essentially preserved under the actionof singular integral operators, such as the Hilbert transform or the Riesz transforms
DECOMPOSITIONS OF GENERALIZED 2-MICROLOCAL BESOV SPACES 23
Rj = −i ∂∂xj
(−∆)−1/2.Moreover, since every classical pseudodifferential operator belonging to S0
1,0 is asum of an operator in Op(Mγ) and a regularizing operator ([13, Chapter 7]), weconclude that the 2-microlocal spaces Bs,mloc
pq (Rn,w) are invariant under the actionof pseudodifferential operators out of S0
1,0.We can state now regularity result for solutions of elliptic partial differential equa-tions. From Calderon and Zygmund [3] we know, that the inverse of an ellipticoperator is a product of a fractional integration and a pseudodifferential operatorof order 0. The latter one is a sum of an operator in Op(Mγ) and a regularizingoperator and the first one is a lift operator (Proposition 2). Using this fact togetherwith the last Theorem, we obtain the following assertion.Theorem 7: Let w ∈ Wα
α1,α2, 0 < p, q ≤ ∞ and s ∈ R. Let Λ be an elliptic partial
differential operator of order m, with smooth coefficients. If Λf = g and g belongsto Bs,mloc
pq (Rn, w), then f belongs to Bs+m,mlocpq (Rn,w).
Appendix
Here we present the more technical proofs of the Lemmas.Proof of Lemma 1. We have to prove that the limes
limr→∞
r∑ν=0
∑
m∈Zn
λνmµνm(x) exists in S ′(Rn). (70)
For ϕ ∈ S(Rn) we get from the moment conditions (17) for fixed ν ∈ N0
∫
Rn
∑
m∈Zn
λνmµνm(y)ϕ(y)dy
=∫
Rn
∑
m∈Zn
λνmµνm(y)wν(y)×
×ϕ(y)−
∑
|β|<L
Dβϕ(2−νm)β!
(y − 2−νm)β
w−1
ν (y)〈y〉κ〈y〉κ dy ,
(71)
where κ > 0 will be specified later on. We use Taylor expansion of ϕ up to theorder L and get with ξ on the line segment joining y and 2−νm
ϕ(y) =∑
|β|<L
Dβϕ(2−νm)β!
(y − 2−νm)β +∑
|β|=L
Dβϕ(ξ)β!
(y − 2−νm)β .
24 HENNING KEMPKA
Using the properties of the weight sequence and 〈y〉κ ≤ 〈y − 2−νm〉κ〈ξ〉κ, we canestimate (71) by
|µνm(y)|∣∣∣∣∣∣ϕ(y)−
∑
|β|<L
Dβϕ(2−νm)β!
(y − 2−νm)β
∣∣∣∣∣∣w−1
ν (y)〈y〉κ〈y〉κ
≤ c2−ν(L−α1)(1 + 2ν |y − 2−νm|)−M×
×∑
|β|=L
|Dβϕ(ξ)|β!
|y − 2−νm|L2νLw−10 (y)
〈ξ〉κ〈y − 2−νm〉κ〈y〉κ
≤ c′2−ν(L−α1)(1 + 2ν |y − 2−νm|)L+κ−M 〈y〉α−κ supξ∈Rn
〈ξ〉κ∑
|β|=L
|Dβϕ(ξ)|β!
.
Hence, we derive from (71)
∣∣∣∣∣∣
∫
Rn
∑
m∈Zn
λνmµνm(y)ϕ(y)dy
∣∣∣∣∣∣≤ c2−ν(L−α1) sup
x∈Rn
〈x〉κ∑
|β|=L
|Dβϕ(x)|β!
×
×∫
Rn
∑
m∈Zn
|λνm|wν(y)(1 + 2ν |y − 2−νm|)L+κ−M 〈y〉α−κdy . (72)
Now, let us suppose that p ≥ 1 then we get applying Holder’s inequality on theintegral in (72) with κ > n
p′ + α ≥ n + α
∫
Rn
∑
m∈Zn
|λνm|wν(y)(1 + 2ν |x− 2−νm|)L+κ−M 〈y〉α−κdy
≤ c2−ν(L+s−α1)2νs
∥∥∥∥∥∑
m∈Zn
|λνm|(1 + 2ν |x− 2−νm|)L+κ−M
∣∣∣∣∣ Lp(Rn, wν)
∥∥∥∥∥ .
By choosing M large enough (M > L + 2n + 2α) and using Lemma 4 with j = νwe get
∣∣∣∣∣∣
∫
Rn
∑
m∈Zn
λνmµνm(y)ϕ(y)dy
∣∣∣∣∣∣
≤ c2−ν(L+s−α1)2ν(s−np )
( ∑
m∈Zn
|λνm|pwpν(2−νm)
)1/p
supx∈Rn
〈x〉κ∑
|β|=L
|Dβϕ(x)|β!
.
Since L > σp − s + α1 = −s + α1 and λ ∈ bs,wpq ↪→ bs,w
p∞ , the convergence of (70) inS ′(Rn) follows.If p < 1, we get analogously by choosing κ = α, M > L+n+2α and using Holder’s
DECOMPOSITIONS OF GENERALIZED 2-MICROLOCAL BESOV SPACES 25
inequality and the weight property wν(y) ≤ Cwν(2−νm)(1 + 2ν |y − 2−νm|)α
∣∣∣∣∣∣
∫
Rn
∑
m∈Zn
λνmµνm(y)ϕ(y)dy
∣∣∣∣∣∣
p
≤ c2−ν(L+s−α1)2νs∑
m∈Zn
|λνm|pwpν(2−νm)
∫
Rn
(1 + 2ν |y − 2−νm|)L+2α−Mdy‖ϕ‖α,L
≤ c′2−ν(L+s+n−np−α1)2ν(s−n
p )∑
m∈Zn
|λνm|pwpν(2−νm)‖ϕ‖α,L .
Finally, using `p ↪→ `1, L > σp − s + α1 = n(
1p − 1
)− s + α1 and λ ∈ bs,w
pq ↪→ bs,wp∞
we get the S ′(Rn) convergence of (70).¤
Proof of Lemma 4. First Step: We treat the case j ≤ ν. We decompose Rn intocubes Qνl and get
‖F |Lp(Rn, wj)‖p =∫
Rn
wpj (x)
∣∣∣∣∣∑
m∈Zn
λνmfνm(x)
∣∣∣∣∣
p
dx
≤ c∑
l∈Zn
∫
Qνl
∣∣∣∣∣∑
m∈Zn
λνmwj(x)(1 + 2j |x− 2−νm|)−R
∣∣∣∣∣
p
dx .
Now, we use
wj(x) ≤ C2α1(ν−j)wν(2−νm)(1 + 2j |x− 2−νm|)α
and that x ∈ Qνl (that means 0 ≤ |x− 2−ν l| ≤ c2−ν). This leads to
‖F |Lp(Rn, wj)‖p
≤ c2α1p(ν−j)∑
l∈Zn
∫
Qνl
∣∣∣∣∣∑
m∈Zn
λνmwν(2−νm)(1 + 2j−ν |l −m|)−R+α
∣∣∣∣∣
p
dx
≤ c2α1p(ν−j)2−νn∑
l∈Zn
( ∑
m∈Zn
|λνm|wν(2−νm)(1 + 2j−ν |l −m|)−R+α
)p
and Young’s inequality gives us
≤ c2α1p(ν−j)2−νn
( ∑
m∈Zn
|λνm|pwpν(2−νm)
)(∑
l∈Zn
(1 + 2j−ν |l|)−R+α
)p
.
(73)
Finally, we have to estimate the last sum in (73). Splitting the sum we obtain∑
l∈Zn
(1 + 2j−ν |l|)−R+α =∑
|l|≤2ν−j
(1 + 2j−ν |l|)−R+α +∑
|l|>2ν−j
(1 + 2j−ν |l|)−R+α
≤ c2(ν−j)n +∫
|y|>2ν−j
(1 + 2j−ν |y|)−R+αdy
≤ c′2(ν−j)n , for R > α + n.
26 HENNING KEMPKA
Putting this into (73) and taking the 1/p power, the first part of the lemma isproved.Second Step: Now, we have j ≥ ν and we can use
fνm(x) = (1 + 2j |x− 2−νm|)−R ≤ c(1 + 2ν |x− 2−νm|)−R
and
wj(x) ≤ C2α2(j−ν)wν(2−νm)(1 + 2ν |x− 2−νm|)α .
The same splitting of the integral in dyadic cubes as in the first step together withthe above inequalities lead to the second case.
¤Proof of Lemma 5. To prove the case ν ≥ j one can modify the steps in [5, Lemma3.3]. We only need the moment conditions on µνm and a sufficiently strong decayof the derivatives of kj,l. The proof for ν ≥ j follows easily by interchanging theroles of kjl and µνm.
¤Proof of Lemma 6. Here we only give the proof for ν ≥ j because the other casefollows by similar arguments. For k ∈ Zn we introduce the quantity
`jν(k) = {m ∈ Zn : Qjk ∩Qνm 6= ∅} = {m ∈ Zn : Qνm ⊆ Qjk} .
Then we get for m ∈ `jν(k)
(1 + |l − k|) ≤ c(1 + 2j |2−j l − 2−νm|) . (74)
Estimate (74) and Minkowski’s inequality yield
(∑
l∈Zn
( ∑
m∈Zn
|bm|(1 + 2j |2−j l − 2−νm|)−R
)p)1/p
=
∑
l∈Zn
∑
k∈Zn
∑
m∈`jν(k)
|bm|(1 + 2j |2−j l − 2−νm|)−R
p
1/p
≤ c
∑
l∈Zn
∑
k∈Zn
(1 + |k − l|)−R∑
m∈`jν(k)
|bm|
p
1/p
= c
∑
l∈Zn
∑
u∈Zn
(1 + |u|)−R∑
m∈`jν(u+l)
|bm|
p
1/p
≤ c∑
u∈Zn
(1 + |u|)−R
∑
l∈Zn
∑
m∈`jν(u+l)
|bm|
p
1/p
.
DECOMPOSITIONS OF GENERALIZED 2-MICROLOCAL BESOV SPACES 27
Finally Holder’s inequality and card `jν(u + l) ∼ 2(ν−j)n imply
(∑
l∈Zn
( ∑
m∈Zn
|bm|(1 + 2j |2−j l − 2−νm|)−R
)p)1/p
≤ c2(ν−j) np′
∑
u∈Zn
(1 + |u|)−R
∑
l∈Zn
∑
m∈`jν(u+l)
|bm|p
1/p
≤ c′2(ν−j) np′
( ∑
m∈Zn
|bm|p)1/p
,
where the last inequality is due to card{l ∈ Zn : Qνm ⊆ Qjl + 2−ju} ∼ 1 andR > n.
¤
References
[1] Andersson, Patrik: Two-microlocal spaces, local norms and weighted spaces. Paper 2 in PhDThesis University of Goteborg (1997), 35-58
[2] Bony, Jean-Michel: Second microlocalization and propagation of singularities for semi-linearhyperbolic equations. Taniguchi Symp. HERT. Katata (1984), 11-49
[3] Calderon, Alberto Pedro; Zygmund, Antoni: Singular integral operators and differentialequations. Amer. J. Math. (1957), 901-921
[4] Edmunds, David Eric; Triebel, Hans: Function spaces, entropy numbers, differential opera-tors. Cambridge Univ. Press (1996)
[5] Frazier, Michael; Jawerth, Bjorn: Decomposition of Besov spaces. Indiana Univ. Math. J. 34(1985), 777-799
[6] Frazier, Michael; Jawerth, Bjorn: A discrete transform and decompositions of distributionspaces. J. Funct. Anal. 93 (1990), 34-170
[7] Frazier, Michael; Jawerth, Bjorn; Weiss, Guido: Littlewood-Paley theory and the study offunction spaces. CBMS Reg. Conf. Ser. Math., AMS, 79 (1991)
[8] Jaffard, Stephane: Pointwise smoothness, two-microlocalisation and wavelet coefficients.Publications Mathematiques 35 (1991), 155-168
[9] Jaffard, Stephane; Meyer, Yves: Wavelet methods for pointwise regularity and local oscilla-tions of functions. Memoirs of the AMS, vol. 123 (1996)
[10] Kempka, Henning: Local characterization of generalized 2-microlocal spaces. Jenaer Schriftenzur Math. & Inf. 20/06 (2006)
[11] Kyriazis, George: Decomposition systems for function spaces. Studia Math. 157 (2003), 133-169
[12] Levy Vehel, Jacques; Seuret, Stephane: The 2-microlocal formalism. Fractal Geometry andApplications: A Jubilee of Benoit Mandelbrot, Proc. Symp. Pure Math., AMS, ProvidenceRI, vol. 72, part2 (2004)
[13] Meyer, Yves, Coifman, Ronald: Wavelets; Calderon-Zygmund and multilinear operators.Cambridge Studies in Advanced Math. , vol. 48 (1997)
[14] Meyer, Yves: Wavelets, vibrations and scalings. CRM monograph series, AMS, vol. 9 (1997)[15] Meyer, Yves; Xu, Hong: Wavelet analysis and chirps. Appl. and Computational Harmonic
Analysis, 4 (1997), 366-379[16] Moritoh, Shinya; Yamada, Tomomi: Two-microlocal Besov spaces and wavelets. Rev. Mat.
Iberoamericana 20 (2004), 277-283[17] Schmeißer, Hans-Jurgen; Triebel, Hans: Topics in Fourier analysis and function spaces.
Leipzig: Akademische Verlagsgesellschaft Geest & Portig (1987)[18] Triebel, Hans: Theory of function spaces. Leipzig: Akademische Verlagsgesellschaft Geest &
Portig (1983)[19] Triebel, Hans: A note on wavelet bases in function spaces. Orlicz Centenary Vol., Banach
Center Publications 64 (2004), 193-206
28 HENNING KEMPKA
[20] Triebel, Hans: Function spaces and wavelets on domains. Zurich: European Math. Soc.Publishing House (2008)
[21] Triebel, Hans: Theory of function spaces III. Basel: Birkhauser (2006)[22] Wojtaszczyk, PrzemysÃlaw: A mathematical introduction to wavelets. Cambridge University
Press, London Math. Society Student Texts 37 (1997)[23] Xu, Hong: Generalisation de la theorie des chirps a divers cadres fonctionnels et application
a leur analyse par ondelettes. Ph. D. thesis, Universite Paris IX Dauphine (1996)
E-mail address: [email protected]
Mathematical Institut, Friedrich Schiller University, Ernst-abbe-Platz 2, 07737Jena